I-NVIDIA yethule isizukulwane esisha samakhadi wezithombe ze-Ampere ngoSepthemba 1, kodwa isethulo sokuqala sasiqukethe cishe imininingwane yobuchwepheshe. Manje, ezinsukwini ezimbalwa kamuva, inkampani ikhiphe imibhalo ecacisa lapho inzuzo yokusebenza ehlaba umxhwele esetha amakhadi wehluzo we-GeForce RTX 30-series ahluke kwawandulelayo.
Abaningi baqaphela ngokushesha ukuthi imininingwane esemthethweni ye-GeForce RTX 3090, i-GeForce RTX 3080 kanye ne-GeForce RTX 3070 kuwebhusayithi ye-NVIDIA ibonise inani elikhulu ngokumangalisayo lamaphrosesa e-CUDA.
Njengoba kwenzeka, ukuphindwa kabili kokusebenza kwe-FP32 kwamaphrosesa egeyimu ye-Ampere uma kuqhathaniswa ne-Turing kuyenzeka ngempela, futhi kuhlotshaniswa noshintsho ekwakhiweni kwamabhlogo okwakha ayisisekelo e-GPU - stream processors (SM).
Ngenkathi ama-SMs kuma-GPU esizukulwane se-Turing ayenendlela eyodwa yokuhlanganisa yokusebenza kwephoyinti elintantayo, e-Ampere iphrosesa yokusakaza ngayinye ithole izindlela ezimbili, ezingakwazi ukwenza imisebenzi efinyelela ku-128 ye-FMA ngomjikelezo wewashi uma kuqhathaniswa no-64 we-Turing. Ngasikhathi sinye, ingxenye yamayunithi okubulala e-Ampere atholakalayo ayakwazi ukwenza kokubili ukusebenza okuphelele (INT) kanye nokusebenza kwephoyinti elintantayo elingu-32-bit (FP32), kuyilapho ingxenye yesibili yamadivayisi ihloselwe ukusebenza kwe-FP32 kuphela. Le ndlela isetshenziselwe ukusindisa isabelomali se-transistor, ngokusekelwe eqinisweni lokuthi umthwalo wegeyimu ukhiqiza kakhulu i-FP32 kunemisebenzi ye-INT. Kodwa-ke, eTuring kwakungekho nhlobo ama-actuator ahlanganisiwe.
Ngesikhathi esifanayo, ukuze kuhlinzekwe amaphrosesa athuthukisiwe okusakaza ngenani elidingekayo ledatha, i-NVIDIA ikhulise usayizi wenqolobane ye-L1 ku-SM ngengxenye yesithathu (kusuka ku-96 kuya ku-128 KB), futhi yaphinda kabili ukuphuma kwayo.
Okunye ukuthuthukiswa okubalulekile ku-Ampere ukuthi ama-CUDA, RT kanye ne-Tensor cores manje angasebenza ngokugcwele ngokuhambisana. Lokhu kuvumela injini yezithombe, isibonelo, ukusebenzisa i-DLSS ukukala uhlaka olulodwa, futhi ngesikhathi esifanayo ibale ifreyimu elandelayo ku-CUDA kanye ne-RT cores, ukunciphisa isikhathi sokuphumula kwamanode asebenzayo nokwandisa ukusebenza okuphelele.
Kulokhu kufanele sengeze ukuthi ama-RT cores esizukulwane sesibili, asetshenziswa e-Amrere, angakwazi ukubala ukuphambana konxantathu abanemisebe ephindwe kabili ngokushesha njengoba kwenzeka e-Turing. Futhi ama-tensor cores amasha esizukulwane sesithathu aphindaphinde ukusebenza kwezibalo lapho esebenza nomatikuletsheni abambalwa.
Ukuphinda kabili isivinini i-Ampere ebala ngaso ukuphambana konxantathu kufanele kube nomthelela omkhulu ekusebenzeni kwezisheshisi ze-GeForce RTX 30-series kumageyimu asekela ukulandela ngomsebe. Ngokusho kwe-NVIDIA, kwakuyilesi sici esisebenza njengebhodlela ekwakhiweni kwe-Turing, kuyilapho isivinini sokubala sokuphambana kwemisebe ye-parallelepipeds ebophayo asizange siphakamise noma yiziphi izikhalazo. Manje ibhalansi yokusebenza ekulandeleni isithuthukisiwe, futhi ngaphezu kwalokho, ku-Ampere, zombili izinhlobo zemisebe ye-ray (ngonxantathu nama-parallelelepipeds) zingenziwa ngokuhambisana.
Ngaphezu kwalokhu, ukusebenza okusha kwengezwe kumakhora we-RT we-Ampere ukuze kuhlanganiswe indawo yonxantathu. Lokhu kungasetshenziswa ukufiphalisa izinto ezinyakazayo uma kungebona bonke onxantathu esigcawini abami ngendlela engashintshi.
Ukukhombisa konke lokhu, i-NVIDIA ibonise ukuqhathanisa okuqondile kokuthi i-Turing ne-Ampere GPUs ikusingatha kanjani ukulandelwa kwe-ray kuWolfenstein Youngblood ekulungisweni kwe-4K. Ngokulandelayo emfanekisweni owethulwe, i-Ampere izuza ngokuphawulekayo ngesivinini sokwakhiwa kohlaka kokubili ngenxa yokubala okusheshayo kwezibalo ze-FP32, ngenxa yezinhlamvu zesizukulwane sesibili ze-RT, kanye nokusebenza okufanayo kwezinsiza ze-GPXNUMX ezihlukahlukene.
Ngaphezu kwalokho, ukuze kuqiniswe lokhu okungenhla, i-NVIDIA yethule imiphumela yokuhlola eyengeziwe ye-GeForce RTX 3090, GeForce RTX 3080 kanye ne-GeForce RTX 3070. Ngokusho kwabo, i-GeForce RTX 3070 icishe ibe ngu-60% ngaphambi kwe-GeForce RTX 2070 ngo-1440p, futhi lesi sithombe sibonwa emidlalweni ngokusekelwa kwe-RTX, kanye ne-rasterization yendabuko, ikakhulukazi e-Borderlands 3.
Ukusebenza kwe-GeForce RTX 3080 kuhle ngokuphindwe kabili kunaleyo ye-GeForce RTX 2080 ekulungisweni kwe-4K. Yiqiniso, kulokhu, e-Borderlands 3 ngaphandle kokusekelwa kwe-RTX, inzuzo yekhadi elisha aliyona kabili, kodwa cishe amaphesenti angu-80.
Futhi ikhadi elidala, i-GeForce RTX 3090, ekuhlolweni kwe-NVIDIA libonisa inzuzo ephindwe kanye nesigamu ngaphezu kwe-Titan RTX.
Ngokwemibiko evela ezintathelini zobuchwepheshe, ukubuyekezwa okugcwele komklamo wereferensi we-GeForce RTX 3080 kufanele kushicilelwe ngoSepthemba 14. Ezinsukwini ezintathu kamuva, ngoSepthemba 17, izovunyelwa ukushicilela idatha yokuhlola yokukhiqiza amamodeli e-GeForce RTX 3080 kubalingani benkampani. Ngakho-ke, sincane kakhulu isikhathi esisele sokulinda imiphumela yokuhlolwa okuzimele yabameleli bochungechunge lwe-GeForce RTX 30 ukuze kuvele ku-inthanethi.
Source:
Source: 3dnews.ru